The Study Of Multi-band Interferometry Aperture Radar Data Inversion Tree Height

Tree height is an essential parameter of the vertical structure of forests,which can assess the biomass and carbon reserves of forests.Tree height inversion was studied at Duke Forest,NC,USA,and Leshan,Sichuan,China.The topography is gentle or with small rolling hills at the Duke,whereas the topography at Leshan is rugged.The inversion algorithms are based on the InSAR and PolInSAR techniques.The ALOS PALSAR and Sentinel-1A SAR datasets are used.1)Particularly,tree heights in the Duke area are estimated at L-band using i)the revised coherent amplitude algorithm,ii)the ESPRIT(Estimation of Signal Parameter via Rotational Invariance Techniques)algorithm,iii)the combined algorithm of i)and ii),and iv)the three-stage algorithm.The tree height is overestimated by i)but underestimate by ii).The combined method improves the estimation.Among the four algorithms,the algorithms iii)and iv)output the best compared to the in situ tree height data.The average heights retrieved by the algorithms are 27.6,6.2,20.0,and 22.3 m,respectively.Among the 32 samples,the number of samples whose absolute difference ratio between the inverted and measured values is less than 20% is 12,1,22,and 24 for the four algorithms,respectively.2)In comparison,algorithms i),ii),and iii)are used to estimate the tree heights using the Sentinel-1A C-band data at Duke Forest.The average heights inverted are22.7,16.3,and 25.4 m,respectively.Among the 15 samples,the number of samples whose absolute difference ratio between the estimated and measured values is less than20% is 11,7,and 3,respectively.Due to the shallow canopy penetration at C-band,the volume scattering dominates.The coherent amplitude algorithm performs the best overall,and the tree height estimation using algorithm ii)is improved.3)The inversion values using the DSM difference method are higher than the measured ones.The improved direct phase difference method underestimates the heights.Then,the two algorithms are combined by the coherence coefficient weights to create another algorithm.Among the 12 sample data,the combined algorithm improves estimations at six cases compared to the two methods.